1 // Protocol Buffers - Google's data interchange format
2 // Copyright 2008 Google Inc. All rights reserved.
3 //
4 // Use of this source code is governed by a BSD-style
5 // license that can be found in the LICENSE file or at
6 // https://developers.google.com/open-source/licenses/bsd
7
8 // Author: laszlocsomor@google.com (Laszlo Csomor)
9 // Based on original Protocol Buffers design by
10 // Sanjay Ghemawat, Jeff Dean, and others.
11
12 // Implementation for long-path-aware open/mkdir/access/etc. on Windows, as well
13 // as for the supporting utility functions.
14 //
15 // These functions convert the input path to an absolute Windows path
16 // with "\\?\" prefix, then pass that to _wopen/_wmkdir/_waccess/etc.
17 // (declared in <io.h>) respectively. This allows working with files/directories
18 // whose paths are longer than MAX_PATH (260 chars).
19 //
20 // This file is only used on Windows, it's empty on other platforms.
21
22 #if defined(_WIN32) && !defined(_XBOX_ONE)
23
24 // Comment this out to fall back to using the ANSI versions (open, mkdir, ...)
25 // instead of the Unicode ones (_wopen, _wmkdir, ...). Doing so can be useful to
26 // debug failing tests if that's caused by the long path support.
27 #define SUPPORT_LONGPATHS
28
29 #include "google/protobuf/io/io_win32.h"
30
31 #include <direct.h>
32 #include <errno.h>
33 #include <fcntl.h>
34 #include <io.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37
38 #include "absl/strings/ascii.h"
39
40 #ifndef WIN32_LEAN_AND_MEAN
41 #define WIN32_LEAN_AND_MEAN 1
42 #endif
43
44 #include <windows.h>
45
46 #include <memory>
47 #include <sstream>
48 #include <string>
49 #include <vector>
50
51 namespace google {
52 namespace protobuf {
53 namespace io {
54 namespace win32 {
55 namespace {
56
57 using std::string;
58 using std::wstring;
59
60 template <typename char_type>
61 struct CharTraits {
62 static bool is_alpha(char_type ch);
63 };
64
65 template <>
66 struct CharTraits<char> {
is_alphagoogle::protobuf::io::win32::__anon59bbff6c0111::CharTraits67 static bool is_alpha(char ch) { return absl::ascii_isalpha(ch); }
68 };
69
70 template <>
71 struct CharTraits<wchar_t> {
is_alphagoogle::protobuf::io::win32::__anon59bbff6c0111::CharTraits72 static bool is_alpha(wchar_t ch) {
73 return (ch >= 'a' && ch <= 'z') || (ch >= 'A' && ch <= 'Z');
74 }
75 };
76
77 template <typename char_type>
null_or_empty(const char_type * s)78 bool null_or_empty(const char_type* s) {
79 return s == nullptr || *s == 0;
80 }
81
82 // Returns true if the path starts with a drive letter, e.g. "c:".
83 // Note that this won't check for the "\" after the drive letter, so this also
84 // returns true for "c:foo" (which is "c:\${PWD}\foo").
85 // This check requires that a path not have a longpath prefix ("\\?\").
86 template <typename char_type>
has_drive_letter(const char_type * ch)87 bool has_drive_letter(const char_type* ch) {
88 return CharTraits<char_type>::is_alpha(ch[0]) && ch[1] == ':';
89 }
90
91 // Returns true if the path starts with a longpath prefix ("\\?\").
92 template <typename char_type>
has_longpath_prefix(const char_type * path)93 bool has_longpath_prefix(const char_type* path) {
94 return path[0] == '\\' && path[1] == '\\' && path[2] == '?' &&
95 path[3] == '\\';
96 }
97
98 template <typename char_type>
is_separator(char_type c)99 bool is_separator(char_type c) {
100 return c == '/' || c == '\\';
101 }
102
103 // Returns true if the path starts with a drive specifier (e.g. "c:\").
104 template <typename char_type>
is_path_absolute(const char_type * path)105 bool is_path_absolute(const char_type* path) {
106 return has_drive_letter(path) && is_separator(path[2]);
107 }
108
109 template <typename char_type>
is_drive_relative(const char_type * path)110 bool is_drive_relative(const char_type* path) {
111 return has_drive_letter(path) && (path[2] == 0 || !is_separator(path[2]));
112 }
113
join_paths(const wstring & path1,const wstring & path2)114 wstring join_paths(const wstring& path1, const wstring& path2) {
115 if (path1.empty() || is_path_absolute(path2.c_str()) ||
116 has_longpath_prefix(path2.c_str())) {
117 return path2;
118 }
119 if (path2.empty()) {
120 return path1;
121 }
122
123 if (is_separator(path1[path1.size() - 1])) {
124 return is_separator(path2[0]) ? (path1 + path2.substr(1))
125 : (path1 + path2);
126 } else {
127 return is_separator(path2[0]) ? (path1 + path2)
128 : (path1 + L'\\' + path2);
129 }
130 }
131
normalize(wstring path)132 wstring normalize(wstring path) {
133 if (has_longpath_prefix(path.c_str())) {
134 path = path.substr(4);
135 }
136
137 static const wstring dot(L".");
138 static const wstring dotdot(L"..");
139 const WCHAR* p = path.c_str();
140
141 std::vector<wstring> segments;
142 int segment_start = -1;
143 // Find the path segments in `path` (separated by "/").
144 for (int i = 0;; ++i) {
145 if (!is_separator(p[i]) && p[i] != L'\0') {
146 // The current character does not end a segment, so start one unless it's
147 // already started.
148 if (segment_start < 0) {
149 segment_start = i;
150 }
151 } else if (segment_start >= 0 && i > segment_start) {
152 // The current character is "/" or "\0", so this ends a segment.
153 // Add that to `segments` if there's anything to add; handle "." and "..".
154 wstring segment(p, segment_start, i - segment_start);
155 segment_start = -1;
156 if (segment == dotdot) {
157 if (!segments.empty() &&
158 (!has_drive_letter(segments[0].c_str()) || segments.size() > 1)) {
159 segments.pop_back();
160 }
161 } else if (segment != dot && !segment.empty()) {
162 segments.push_back(segment);
163 }
164 }
165 if (p[i] == L'\0') {
166 break;
167 }
168 }
169
170 // Handle the case when `path` is just a drive specifier (or some degenerate
171 // form of it, e.g. "c:\..").
172 if (segments.size() == 1 && segments[0].size() == 2 &&
173 has_drive_letter(segments[0].c_str())) {
174 return segments[0] + L'\\';
175 }
176
177 // Join all segments.
178 bool first = true;
179 std::wstringstream result;
180 for (int i = 0; i < segments.size(); ++i) {
181 if (!first) {
182 result << L'\\';
183 }
184 first = false;
185 result << segments[i];
186 }
187 // Preserve trailing separator if the input contained it.
188 if (!path.empty() && is_separator(p[path.size() - 1])) {
189 result << L'\\';
190 }
191 return result.str();
192 }
193
as_windows_path(const char * path,wstring * result)194 bool as_windows_path(const char* path, wstring* result) {
195 if (null_or_empty(path)) {
196 result->clear();
197 return true;
198 }
199 wstring wpath;
200 if (!strings::utf8_to_wcs(path, &wpath)) {
201 return false;
202 }
203 if (has_longpath_prefix(wpath.c_str())) {
204 *result = wpath;
205 return true;
206 }
207 if (is_separator(path[0]) || is_drive_relative(path)) {
208 return false;
209 }
210
211
212 if (!is_path_absolute(wpath.c_str())) {
213 int size = ::GetCurrentDirectoryW(0, nullptr);
214 if (size == 0 && GetLastError() != ERROR_INSUFFICIENT_BUFFER) {
215 return false;
216 }
217 std::unique_ptr<WCHAR[]> wcwd(new WCHAR[size]);
218 ::GetCurrentDirectoryW(size, wcwd.get());
219 wpath = join_paths(wcwd.get(), wpath);
220 }
221 wpath = normalize(wpath);
222 if (!has_longpath_prefix(wpath.c_str())) {
223 // Add the "\\?\" prefix unconditionally. This way we prevent the Win32 API
224 // from processing the path and "helpfully" removing trailing dots from the
225 // path, for example.
226 // See https://github.com/bazelbuild/bazel/issues/2935
227 wpath = wstring(L"\\\\?\\") + wpath;
228 }
229 *result = wpath;
230 return true;
231 }
232
233 } // namespace
234
open(const char * path,int flags,int mode)235 int open(const char* path, int flags, int mode) {
236 #ifdef SUPPORT_LONGPATHS
237 wstring wpath;
238 if (!as_windows_path(path, &wpath)) {
239 errno = ENOENT;
240 return -1;
241 }
242 return ::_wopen(wpath.c_str(), flags, mode);
243 #else
244 return ::_open(path, flags, mode);
245 #endif
246 }
247
mkdir(const char * path,int)248 int mkdir(const char* path, int /*_mode*/) {
249 #ifdef SUPPORT_LONGPATHS
250 wstring wpath;
251 if (!as_windows_path(path, &wpath)) {
252 errno = ENOENT;
253 return -1;
254 }
255 return ::_wmkdir(wpath.c_str());
256 #else // not SUPPORT_LONGPATHS
257 return ::_mkdir(path);
258 #endif // not SUPPORT_LONGPATHS
259 }
260
access(const char * path,int mode)261 int access(const char* path, int mode) {
262 #ifdef SUPPORT_LONGPATHS
263 wstring wpath;
264 if (!as_windows_path(path, &wpath)) {
265 errno = ENOENT;
266 return -1;
267 }
268 return ::_waccess(wpath.c_str(), mode);
269 #else
270 return ::_access(path, mode);
271 #endif
272 }
273
chdir(const char * path)274 int chdir(const char* path) {
275 #ifdef SUPPORT_LONGPATHS
276 wstring wpath;
277 if (!as_windows_path(path, &wpath)) {
278 errno = ENOENT;
279 return -1;
280 }
281 return ::_wchdir(wpath.c_str());
282 #else
283 return ::_chdir(path);
284 #endif
285 }
286
stat(const char * path,struct _stat * buffer)287 int stat(const char* path, struct _stat* buffer) {
288 #ifdef SUPPORT_LONGPATHS
289 wstring wpath;
290 if (!as_windows_path(path, &wpath)) {
291 errno = ENOENT;
292 return -1;
293 }
294 return ::_wstat(wpath.c_str(), buffer);
295 #else // not SUPPORT_LONGPATHS
296 return ::_stat(path, buffer);
297 #endif // not SUPPORT_LONGPATHS
298 }
299
fopen(const char * path,const char * mode)300 FILE* fopen(const char* path, const char* mode) {
301 #ifdef SUPPORT_LONGPATHS
302 if (null_or_empty(path)) {
303 errno = EINVAL;
304 return nullptr;
305 }
306 wstring wpath;
307 if (!as_windows_path(path, &wpath)) {
308 errno = ENOENT;
309 return nullptr;
310 }
311 wstring wmode;
312 if (!strings::utf8_to_wcs(mode, &wmode)) {
313 errno = EINVAL;
314 return nullptr;
315 }
316 return ::_wfopen(wpath.c_str(), wmode.c_str());
317 #else
318 return ::fopen(path, mode);
319 #endif
320 }
321
close(int fd)322 int close(int fd) { return ::_close(fd); }
323
dup(int fd)324 int dup(int fd) { return ::_dup(fd); }
325
dup2(int fd1,int fd2)326 int dup2(int fd1, int fd2) { return ::_dup2(fd1, fd2); }
327
read(int fd,void * buffer,size_t size)328 int read(int fd, void* buffer, size_t size) {
329 return ::_read(fd, buffer, size);
330 }
331
setmode(int fd,int mode)332 int setmode(int fd, int mode) { return ::_setmode(fd, mode); }
333
write(int fd,const void * buffer,size_t size)334 int write(int fd, const void* buffer, size_t size) {
335 return ::_write(fd, buffer, size);
336 }
337
testonly_utf8_to_winpath(const char * path)338 wstring testonly_utf8_to_winpath(const char* path) {
339 wstring wpath;
340 return as_windows_path(path, &wpath) ? wpath : wstring();
341 }
342
ExpandWildcards(const string & path,std::function<void (const string &)> consume)343 ExpandWildcardsResult ExpandWildcards(
344 const string& path, std::function<void(const string&)> consume) {
345 if (path.find_first_of("*?") == string::npos) {
346 // There are no wildcards in the path, we don't need to expand it.
347 consume(path);
348 return ExpandWildcardsResult::kSuccess;
349 }
350
351 wstring wpath;
352 if (!as_windows_path(path.c_str(), &wpath)) {
353 return ExpandWildcardsResult::kErrorInputPathConversion;
354 }
355
356 static const wstring kDot = L".";
357 static const wstring kDotDot = L"..";
358 WIN32_FIND_DATAW metadata;
359 HANDLE handle = ::FindFirstFileW(wpath.c_str(), &metadata);
360 if (handle == INVALID_HANDLE_VALUE) {
361 // The pattern does not match any files (or directories).
362 return ExpandWildcardsResult::kErrorNoMatchingFile;
363 }
364
365 string::size_type pos = path.find_last_of("\\/");
366 string dirname;
367 if (pos != string::npos) {
368 dirname = path.substr(0, pos + 1);
369 }
370
371 ExpandWildcardsResult matched = ExpandWildcardsResult::kErrorNoMatchingFile;
372 do {
373 // Ignore ".", "..", and directories.
374 if ((metadata.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) == 0 &&
375 kDot != metadata.cFileName && kDotDot != metadata.cFileName) {
376 matched = ExpandWildcardsResult::kSuccess;
377 string filename;
378 if (!strings::wcs_to_utf8(metadata.cFileName, &filename)) {
379 matched = ExpandWildcardsResult::kErrorOutputPathConversion;
380 break;
381 }
382
383 if (dirname.empty()) {
384 consume(filename);
385 } else {
386 consume(dirname + filename);
387 }
388 }
389 } while (::FindNextFileW(handle, &metadata));
390 FindClose(handle);
391 return matched;
392 }
393
394 namespace strings {
395
wcs_to_mbs(const WCHAR * s,string * out,bool outUtf8)396 bool wcs_to_mbs(const WCHAR* s, string* out, bool outUtf8) {
397 if (null_or_empty(s)) {
398 out->clear();
399 return true;
400 }
401 BOOL usedDefaultChar = FALSE;
402 SetLastError(0);
403 int size = WideCharToMultiByte(
404 outUtf8 ? CP_UTF8 : CP_ACP, 0, s, -1, nullptr, 0, nullptr,
405 outUtf8 ? nullptr : &usedDefaultChar);
406 if ((size == 0 && GetLastError() != ERROR_INSUFFICIENT_BUFFER)
407 || usedDefaultChar) {
408 return false;
409 }
410 std::unique_ptr<CHAR[]> astr(new CHAR[size]);
411 WideCharToMultiByte(
412 outUtf8 ? CP_UTF8 : CP_ACP, 0, s, -1, astr.get(), size, nullptr, nullptr);
413 out->assign(astr.get());
414 return true;
415 }
416
mbs_to_wcs(const char * s,wstring * out,bool inUtf8)417 bool mbs_to_wcs(const char* s, wstring* out, bool inUtf8) {
418 if (null_or_empty(s)) {
419 out->clear();
420 return true;
421 }
422
423 SetLastError(0);
424 int size =
425 MultiByteToWideChar(inUtf8 ? CP_UTF8 : CP_ACP, 0, s, -1, nullptr, 0);
426 if (size == 0 && GetLastError() != ERROR_INSUFFICIENT_BUFFER) {
427 return false;
428 }
429 std::unique_ptr<WCHAR[]> wstr(new WCHAR[size]);
430 MultiByteToWideChar(
431 inUtf8 ? CP_UTF8 : CP_ACP, 0, s, -1, wstr.get(), size + 1);
432 out->assign(wstr.get());
433 return true;
434 }
435
utf8_to_wcs(const char * input,wstring * out)436 bool utf8_to_wcs(const char* input, wstring* out) {
437 return mbs_to_wcs(input, out, true);
438 }
439
wcs_to_utf8(const wchar_t * input,string * out)440 bool wcs_to_utf8(const wchar_t* input, string* out) {
441 return wcs_to_mbs(input, out, true);
442 }
443
444 } // namespace strings
445 } // namespace win32
446 } // namespace io
447 } // namespace protobuf
448 } // namespace google
449
450 #endif // defined(_WIN32)
451